Google Cloud Unveils New AI Tools for Agentic Architects

Google has released a wave of new AI-native tools and features aimed at equipping “agentic architects” to move beyond analysis and into action.

The new capabilities, announced at Big Data London, promise to eliminate long-standing productivity roadblocks while giving data scientists the power to build real-world agents that can operate at enterprise scale.

Breaking down friction in data science

Yasmeen Ahmad, Managing Director, Data Cloud, Google Cloud, explains that one of the biggest challenges for data scientists has long been the friction of fragmented workflows — toggling between SQL clients, Python notebooks, Spark clusters, and visualization tools. To address this, Google introduced fundamental upgrades to Colab Enterprise notebooks in BigQuery and Vertex AI, including:

• Native SQL cells: Combine SQL and Python within the same notebook.
• Interactive visualization cells: Automatically generate editable charts to speed up analysis.

These features unify exploration, modeling, and visualization into a single environment, creating an integrated development experience for data science.

Complementing this, Google’s Data Science Agent acts as an “interactive partner” within Colab, now with enhanced tool usage including BigQuery ML, BigQuery DataFrames, and Spark. Alongside, the Lightning Engine, now generally available, promises a 4x boost to Spark performance and full compatibility with ML and AI workloads.

Building agents that understand the real world

To power intelligent agents, data scientists need access to real-time and unstructured data. Google unveiled several key innovations:

• Stateful processing for BigQuery continuous queries: SQL queries can now incorporate “memory,” enabling advanced pattern detection in live data, such as spotting suspicious spending behavior in real time.
• Autonomous embedding generation in BigQuery: A boost for AI applications, eliminating the need for custom pipelines by automatically generating vector embeddings over multimodal data.

These are already in use. For example, UK supermarket chain Morrisons uses Google’s vector capabilities to power its in-store product finder, handling 50,000 daily searches and guiding customers in real time.

From notebook to production

Recognizing that building an agent is only the beginning, Ahmad says Google introduced the Agent Development Kit, enabling teams to orchestrate fleets of production-grade agents. These agents can not only detect issues but also take autonomous actions, such as logging cases in ServiceNow or Salesforce.

Connectivity — long a pain point for enterprises — is also addressed. With first-party BigQuery tools and the MCP Toolbox, agent fleets can integrate across Google Cloud data platforms like BigQuery, AlloyDB, Cloud SQL, and Spanner.

Meanwhile, developers themselves can use new Gemini CLI extensions for natural language data tasks directly in the terminal, reducing the need for UI-based workflows.

Architecting the future

Ahmad reckons these innovations collectively shift the data scientist’s role from analyst to architect, empowering them to engineer systems that sense, reason, and act with intelligence. With an AI-native stack unifying the environment, broadening access to real-world data, and enabling production-ready multi-agent systems, the future of data science is potentially being rewritten.

Google is inviting organizations to explore these breakthroughs with its newly released Data Science eBook, featuring eight use cases to help teams get started.